Explainable Artificial Intelligence for Ovarian Cancer: Biomarker Contributions in Ensemble Models

Objectives: To assess the diagnostic accuracy of different risk of malignancy index (RMI) scores and to evaluate the role of a modified RMI (RMI 5) in pre-operative discrimination between benign and malignant ovarian masses. Study Design: Prospective observational study. Patients and methods: Women with a suspicious ovarian mass scheduled for laparotomy or laparoscopy were potentially eligible for inclusion in the current study. Trans-abdominal and trans-vaginal ultrasound with Doppler assessment of the adnexal masses was done. Calculation of the RMI 1, RMI2, RMI 3, RMI 4, and RMI5 was done. We compared RMI to histopathological outcome. In the current study, a new RMI score was created by adding Doppler blood flow of the ovarian mass to the calculation of the previous RMI1. Results: One hundred and fifty women with ovarian masses were included in the current study. Ninety six women (64%) had benign ovarian masses while, malignant ovarian masses were found in 54 women (36%). Comparison between benign and malignant ovarian masses regarding to the risk of malignancy indices revealed that that there was a statistically significant difference between the two groups regarding to the risk of malignancy indices with Pvalue< 0.001. Receiver operator characteristic curve analysis of the 5 RMI indices shows that the best method for prediction of malignant ovarian tumor was RMI 1. Also there was no statistically significant difference between the five methods in prediction of malignant ovarian tumors. RMI5 with cut off value of 250 is reliable tool at a tertiarycenter to discriminate between ovarian cancer and benign ovarian masses with a sensitivity of 90.38% and specificity of 93.88%. There was statistically significant difference between the different stages of ovarian cancer and RMI 5 with (P<0.05). Conclusion: The RMI 1 is the gold standard for preoperative discrimination between benign and malignant ovarian masses. Adding Doppler flow to the parameters of RMI 1 (RMI 5) increased specificity of RMI 1 in detecting malignant ovarian masses.

BACKGROUND Ovarian cancer (OC) is the most lethal gynecological cancer among females, and its early diagnosis could help for better outcomes of the patients. AIM To investigate the utility of serum insulin-like growth factors-binding proteins 2 (IGFBP2), secreted phosphoprotein 1 (SPP1), thrombospondin 1 protein (TSP1) and D-dimer levels in addition to currently used biomarkers [cancer antigen 125 (CA125) and human epididymis protein 4 (HE4)] in the diagnosis of epithelial OC (EOC). METHODS This is a case-control study that included fifty females diagnosed with EOC, 10 females with benign ovarian masses recruited from the Egyptian National Cancer Institute, and 30 healthy females as a control group. All subjects were assessed for serum HE4, CA125, IGFBP2, TSP1 and SPP1 measurement by enzyme-linkedimmunosorbent assay. RESULTS There was a statistically significant difference in serum levels between EOC, benign ovarian masses, and healthy control groups regarding CA125 and SPP1 (P &lt; 0.001 for both markers), while HE4 and IGFBP2 increased significantly in EOC compared to healthy control groups (P &lt; 0.001 for all markers) with no significant difference between EOC and benign ovarian masses groups. However, there was no statistically significant difference among EOC, benign ovarian masses, and healthy control groups regarding the TSP1 serum levels (P = 0.051). Receiver operating characteristic analysis revealed that combined assessment of SPP1 with CA125 or TSP1 increased the diagnosis of EOC patients to a sensitivity, specificity, and area under curve of (93.3%, 100%, 0.968; respectively, P &lt; 0.001). CONCLUSION SPP1 may be a potential marker for the differentiation between benign and malignant ovarian masses, while IGFBP2 can differentiate between healthy females and females with ovarian masses. Combining SPP1 with CA125 or TSP1 provides high sensitivity and specificity for the detection of EOC patients.

Ovarian cancer is associated with high morbidity and mortality. This is due to the nonspecific symptoms and no effective screening methods. Currently, carbohydrate antigen-125 (CA125) is used as a tumor biomarker for the diagnosis of ovarian cancer, but it has its own limitations. Hence, there is a need for other tumor biomarkers for the diagnosis of ovarian cancer. Objective of the study was to evaluate the diagnostic test characteristics of plasma osteopontin (OPN) in detecting ovarian malignancy and comparing its performance with CA125. This is a prospective cross-sectional diagnostic test evaluation. Women with adnexal mass detected by clinical or radiological examination were enrolled as suspected cases. Women who presented with other gynecological conditions were enrolled as controls. OPN and CA125 levels were measured in all enrolled subjects. Among 106 women enrolled, 26 were ovarian cancer, 31 had benign ovarian masses, and 49 were controls. Median plasma CA125 levels were higher in subjects with ovarian cancer (298 U/ml; interquartile range [IQR]: 84-1082 U/ml vs. 37.5U/ml; IQR: 17.6-82.9U/ml; P < 0.001). CA125 sensitivity, specificity, positive, and negative likelihood ratios were 88.5%, 61.3%, 2.10, and 0.19, respectively. Median plasma OPN levels were higher in subjects with ovarian cancer (63.1 ng/ml; IQR: 39.3-137 ng/ml vs. 27 ng/ml; IQR: 20-52 ng/ml; P = 0.001). Sensitivity, specificity, positive, and negative likelihood ratios of OPN were 50%, 87%, 2.58, and 0.62, respectively. OPN levels were higher in ovarian cancer than in the benign ovarian mass and had better specificity than CA125. OPN can better differentiate between benign and malignant ovarian mass as compared to CA125.

Ovarian cancer is a common female malignancy frequently identified at advanced stages. Diffusion-weighted imaging (DWI) provides valuable information on structural traits of tissue and is used as an imaging biomarker in OST cancer prognosis. Post-processing of three-dimensional apparent diffusion coefficient (ADC) maps has proven useful in evaluating variable tumors, although its position in ovarian cancer prognosis is until now not well defined. Consequently, our foremost objective was to assess the sensitivity and efficiency of DWI (T1 and T2) and ADC maps in malignant and benign ovarian lesions prognosis. A total of 58 patients with undetermined ovarian masses in ultrasound were referred to MRI for more accurate diagnosis. The signals of DWI (qualitative) and ADC values (quantitative DWI) of the lesion components were analyzed separately. Student's t-test and receiver operating characteristic (ROC) curves were used to determine the ability of DWI and ADC in the discrimination between malignant and benign ovarian masses. Of the 58 masses, 33 have been benign, and 25 have been malignant. There was a decrease correlation between signal thing on T2W and ADC values in malignant as compared to benign masses. The DWI and T1 + GAD values in malignant tumors have been substantially higher than the ones in benign masses (P value < 0.0001). Additionally, our consequences suggest that a T1 cutoff value (1 × 10⁻≥ mm²/s) would possibly the quality factor to help discriminate between benign and malignant lesions. The mixture of DWI imaging with T1 + GAD values can beautify the diagnostic overall performance in discrimination among benign and malignant ovarian masses by increasing specificity.

BackgroundCancer antigen 125 (CA125) and human epididymis protein 4 (HE4) are the most commonly used tumor biomarkers for ovarian cancer (OC) screening and diagnosis. The risk of ovarian malignancy algorithm (ROMA) score uses these markers, as detected by the Roche system, to predict the risk of OC. This study sought to assess the performance of the Mindray system in detecting CA125 and HE4 for ROMA score calculation in clinical settings.MethodsConsecutive OC patients and patients with benign pelvic masses were screened and enrolled in this study. The CA125 and HE4 levels of these patients were measured using both the Mindray and Roche systems. The ROMA score for each patient was calculated. Diagnostic performance was evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve.ResultsThe HE4 and CA125 levels were significantly higher in the patients with OC than the patients with benign ovarian masses. Both detection systems showed high efficiency in detecting ovarian cancer. For the premenopausal OC patients, the AUC values for the ROMA score, HE4, and CA125 were 0.866, 0.852, and 0.879, respectively, using the Roche system, and 0.911, 0.902, and 0.883, respectively, using the Mindray system. For the postmenopausal OC patients, the AUC values for the ROMA score, HE4, and CA125 were 0.962, 0.920, and 0.953, respectively, using Roche system, and 0.966, 0.924, and 0.959, respectively, using the Mindray system. The correlation analysis showed strong agreement between the two systems. Among the patients who experienced recurrence, we observed a significant increase in both HE4 and CA125 levels compared to baseline using the Mindray system.ConclusionsThe Mindray and Roche systems provide consistent results. The Mindray system can be used to detect HE4 and CA125 for ROMA score calculation.

Ovarian cancer remains a highly lethal disease due to its late clinical presentation and lack of reliable early biomarkers. Protein-based diagnostic markers have presented limitations in identifying ovarian cancer. We tested the potential of phospholipids as markers of ovarian cancer by utilizing inter-related regulation of phospholipids, a unique property that allows the use of ratios between phospholipid species for quantitation. High-performance liquid chromatography mass spectrometry was used to measure phospholipid, lysophospholipid, and sphingophospholipid content in plasma from patients with benign ovarian masses, patients with ovarian cancer, and controls. We applied both absolute and relative phospholipid ratios for quantitation. Receiver operating characteristic analysis was performed to test the sensitivity and specificity. We found that utilization of ratios between phospholipid species greatly outperformed absolute quantitation in the identification of ovarian cancer. Of the phospholipids analyzed, species in phosphatidylcholine (PC), lysophosphatidylcholine (LPC), and sphingomyelin (SM) were found to have great biomarker potential. LPC(20:4)/LPC(18:0) carried the greatest capacity to differentiate cancer from control, SM(d18:1/24:1)/SM(d18:1/22:0) to differentiate benign from cancer, and PC(18:0/20:4)/PC(18:0/18:1) to differentiate benign from control. These results demonstrate the potential of plasma phospholipids as a novel marker of ovarian cancer by utilizing the unique characteristics of phospholipids to further enhance the diagnostic power.

Commentary on 'Performance of ultrasound as a second line test to serum CA125 in ovarian cancer screening'.

Background: Risk of ovarian malignancy algorithm (ROMA) combining human epididymis secretory protein 4 (HE4) and CA125 is a novel score, specific for epithelial ovarian cancer (EOC). Method: Our cohort prospective study aimed to evaluate the role of HE4 and ROMA score in the diagnosis of EOC. We determined CA125 and HE4 serum levels in 56 premenopausal women with ovarian mass (38 women with benign ovarian mass and 18 women with malignant ovarian mass), 56 postmenopausal women with ovarian mass (20 women with benign ovarian mass and 36 women with malignant ovarian mass), and 56 healthy women as control. Results: Serum CA125 and HE4 and ROMA score were significantly higher among postmenopausal group compared with premenopausal and control groups (P< 0.001), and the median serum CA125 and HE4 and ROMA levels were statistically higher among malignant lesions compared with benign lesions and control group (P< 0.001). The sensitivity and specificity of HE4 and ROMA vs. CA125 in discriminating ovarian cancer from benign ovarian tumor was (88% and 98% vs. 90%) and (97% and 99% vs. 80%), respectively. ROMA had better sensitivity and specificity compared to CA125 and HE4 in premenopausal and postmenopausal women (P <0.001) In premenopausal patients, there was a statistically significant difference regarding the area under the curve (AUC) of ROMA vs. CA125 (P=0.004) and ROMA vs. HE4 (P =0.02). Conclusion: ROMA score showed a better performance in comparison with either CA125 or HE4 alone in premenopausal patients. HE4 and ROMA score significantly differentiated early from late stage ovarian cancer.

Introduction: Ovarian cancer (OC) is one of the deadliest cancers, with 314,000 new cases and 207,000 deaths globally in 2020. Serum CA125 has been explored as an OC biomarker for the past 40 years, but lacks sensitivity for early stage OC and is not recommended for screening average-risk, asymptomatic women. We hypothesize that co-localization of biomarkers on the surface of individual extracellular vesicles (EVs), which are shed into the circulation by cancer cells, may lead to development of a blood test for early stage OC. We evaluated the potential of our approach in detecting early stage OC in clinical samples. Methods: We isolated EVs using size-exclusion chromatography and immunoaffinity capture, and detected biomarkers co-localized on the surface of individual EVs with proximity ligation qPCR. Using this approach, we evaluated 49 antibody combinations recognizing 2 or more biomarkers. Each combination consisted of 1 capture antibody and 2 oligonucleotide-tagged detection antibodies. We tested plasma samples from women with early stage I/II high-grade serous ovarian carcinoma (HGSOC)(n=18; 48-80 yr, med 57) and late stage HGSOC (n=24; 37-80 yr, med 54). HGSOC samples were sourced from 2 commercial vendors. Controls comprised samples from women with benign ovarian masses (n=26; 23-76 yr, med 39.5) sourced from a single vendor, and samples prospectively collected by Mercy from healthy women with no cancer history (n=24; 22-72 yr, med 52.5). PCR cycle threshold (Ct) values were measured for each of 49 combinations and data was evaluated using univariate analysis. Performance was compared to plasma CA125 measured at Mercy by commercial ELISA. Results: 8 of 49 combinations distinguished all stages of HGSOC relative to benign and healthy controls with AUCs ranging from 0.86 (95% CI 0.78-0.94) to 0.95 (95% CI 0.90-1.00), comparable to CA125 with an AUC of 0.87 (95% CI 0.79-0.95). One of the most effective combinations (STn, BST2, MUC1) had a sensitivity of 0.78 (95% CI 0.52-0.94) at a specificity of 0.96 (95% CI 0.87-0.99) in detecting early stage HGSOC. This combination also detected HGSOC in 6 of 11 women (3 early stage, 3 late stage) with normal CA125 (&amp;lt; 25 U/mL) and correctly classified 7 of 8 women with benign masses and high CA125 (&amp;gt; 25 U/mL). Conclusions: These preliminary data suggest that co-localization of surface biomarkers in single EVs may provide an effective means to identify women with early stage HGSOC, including those with normal CA125, while avoiding false positives in women with benign masses and high CA125. Despite the inherent challenges associated with commercial samples, our finding that several combinations detected early stage HGSOC is promising. Statistically powered studies with curated repository specimens are underway to refine combinations and independently validate our assay for early stage OC detection. Citation Format: Laura T. Bortolin, Daniel P. Salem, Sanchari Banerjee, Kelly M. Biette, Delaney M. Byrne, Anthony D. Couvillon, Peter A. Duff, Jonian Grosha, MacKenzie Sadie King, Christopher R. Sedlak, Ibukunoluwapo O. Zabroski, Claire Alexander, Karen Copeland, Daniel Gusenleitner, Emily S. Winn-Deen, Eric K. Huang, Bo R. Rueda, Joseph Charles Sedlak. Preliminary results for a novel single extracellular vesicle assay for early stage ovarian cancer: The power of co-localized detection of surface biomarkers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 3390.

BackgroundEndometriosis is frequently associated with high levels of CA125. This marker is therefore not useful for discriminating ovarian endometrioma from ovarian malignancy. The aim of this study was to establish a panel of complementary biomarkers that could be helpful in the differential diagnosis between ovarian endometriosis or other ovarian benign masses and ovarian cancer.MethodsBlood samples from 50 healthy women, 17 patients with benign ovarian tumors, 57 patients with ovarian endometrioma and 39 patients with ovarian cancer were analyzed and serum values were measured for the following biomarkers: CA125, HE4 and CA72-4.ResultsSerum CA125 concentration was elevated in both patients with ovarian endometriosis and ovarian cancer but not in patients with other benign ovarian masses. HE4 was never increased in patients with endometriosis or benign masses whereas it was significantly higher in all patients with ovarian cancer (p < 0.05). A marked difference in CA72-4 values was observed between women with ovarian cancer (67%) and those with endometriosis (p < 0.05).ConclusionsThe results of the study suggest that HE4 and CA72-4 determination is the best approach to confirm the benign nature of ovarian endometrioma in women with high CA125 levels.

The United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKCTOCS) recently reported a reduction in the average overall mortality among ovarian cancer patients screened with an annual sequential, multimodal strategy that tracked biomarker CA125 over time, where increasing serum CA125 levels prompted ultrasound. However, multiple cases were documented wherein serum CA125 levels were rising, but ultrasound screens were normal, thus delaying surgical intervention. A significant factor which could contribute to false negatives is that many aggressive ovarian cancers are believed to arise from epithelial cells on the fimbriae of the fallopian tubes, which are not readily imaged. Moreover, because only a fraction of metastatic tumors may reach a sonographically-detectable size before they metastasize, annual screening with ultrasound may fail to detect a large fraction of early-stage ovarian cancers. The ability to detect ovarian carcinomas before they metastasize is critical and future efforts towards improving screening should focus on identifying unique features specific to aggressive, early-stage tumors, as well as improving imaging sensitivity to allow for detection of tubal lesions. Implementation of a three-stage multimodal screening strategy in which a third modality is employed in cases where the first-line blood-based assay is positive and the second-line ultrasound exam is negative may also prove fruitful in detecting early-stage cases missed by ultrasound.

Background:Ovarian cancer is the seventh most common cancer in females with the highest mortality rate of all gynecological cancers due to its late discovery and ambiguous symptoms. Thus, there is a need for new promising strategies to diagnose ovarian cancer. We aimed at finding a characteristic plasma proteome pattern that could be used for the detection of epithelial ovarian cancer, in comparison with benign ovarian masses and healthy controls. We also aimed at differentiating between profiling of plasma proteins in early and advanced stages of ovarian cancer and between serous and non-serous histopathological types. Methods:The combination of MagSi-proteomics C8 beads, Ultraflextreme MALDI-TOF and ClinPro Tools software was used to compare the plasma protein spectra from 50 patients with epithelial ovarian cancer, 20 patients with benign ovarian masses and 50 age matched healthy females. Results:A plasma proteome profile of 21 peaks differentiated patients with epithelial ovarian cancer from healthy controls with a sensitivity of 73 % and a specificity of 82.8% upon external validation, while a 5-peak profile differentiated patients with epithelial ovarian cancer from patients with benign ovarian masses with a sensitivity of 81% and a specificity of 73.7%. A 20 peak profile was generated to discriminate between early and late stages of the disease with 88.3% recognition capability and 70% cross validation. Conclusion:MALDI-TOF proteomic profiling represents a promising potential tool for diagnosing epithelial ovarian cancer, discriminating between early and advanced stages and between serous and non- serous types.

Background:Early diagnosis of ovarian cancer is essential for long term disease control and mortality reduction. This has been achieved using tumor markers like cancer antigen 125 (CA-125) which is elevated in malignant as well as non-malignant conditions. This dilemma led to efforts towards development of newer markers like serum human epididymis secretory protein E4 (HE4). Present study aimed to evaluate role of HE4 in diagnosing ovarian cancers and comparing it with CA-125. Methods:Serum samples from 67 patients with ovarian cancer, 42 with benign ovarian masses and 26 healthy controls were collected preoperatively and tested for serum HE4 levels and CA-125 levels. Diagnostic performance of both tumor markers (HE4/CA-125) to diagnose malignancy in ovarian masses was calculated and compared to each other. Results:Mean CA-125 and HE4 levels were significantly higher in patients with ovarian cancer than in those with benign disease (p<0.001) or healthy controls (p< 0.001). Serum HE4 levels significantly increased in epithelial ovarian cancers when compared to non-epithelial ovarian cancers (p<0.01). Using benign control as comparison, receiver operating characteristic curve (ROC) was generated to predict a cut-off value for diagnosing malignancy for serum HE4 and CA-125. Compared to CA-125, HE4 had a similar sensitivity (83.6% vs. 85.10%) and higher specificity (100% vs. 90.48%); combination of serum HE4 and CA-125 improved the sensitivity to detect ovarian cancer to 92.54%. Sensitivity of HE4 to detect early stage ovarian cancer was superior to CA-125 (92.61% vs. 63.41%). Conclusion:Serum HE4, a novel tumor marker, discriminated epithelial ovarian cancer from benign ovarian masses. HE4 levels were related to the stage and histological types with the lowest levels in mucinous epithelial ovarian cancer and non-epithelial malignancy. Measuring serum HE4 levels alongwith CA-125 may provide higher accuracy for detecting epithelial ovarian cancer particularly in the early stages.

BackgroundHigh ovarian cancer mortality rates motivate the development of effective and patient-friendly diagnostics. Here, we explored the potential of molecular testing in patient-friendly samples for ovarian cancer detection.MethodsHome-collected urine, cervicovaginal self-samples, and clinician-taken cervical scrapes were prospectively collected from 54 patients diagnosed with a highly suspicious ovarian mass (benign n = 25, malignant n = 29). All samples were tested for nine methylation markers, using quantitative methylation-specific PCRs that were verified on ovarian tissue samples, and compared to non-paired patient-friendly samples of 110 age-matched healthy controls. Copy number analysis was performed on a subset of urine samples of ovarian cancer patients by shallow whole-genome sequencing.ResultsThree methylation markers are significantly elevated in full void urine of ovarian cancer patients as compared to healthy controls (C2CD4D, P = 0.008; CDO1, P = 0.022; MAL, P = 0.008), of which two are also discriminatory in cervical scrapes (C2CD4D, P = 0.001; CDO1, P = 0.004). When comparing benign and malignant ovarian masses, GHSR shows significantly elevated methylation levels in the urine sediment of ovarian cancer patients (P = 0.024). Other methylation markers demonstrate comparably high methylation levels in benign and malignant ovarian masses. Cervicovaginal self-samples show no elevated methylation levels in patients with ovarian masses as compared to healthy controls. Copy number changes are identified in 4 out of 23 urine samples of ovarian cancer patients.ConclusionsOur study reveals increased methylation levels of ovarian cancer-associated genes and copy number aberrations in the urine of ovarian cancer patients. Our findings support continued research into urine biomarkers for ovarian cancer detection and highlight the importance of including benign ovarian masses in future studies to develop a clinically useful test.

To apply the International Ovarian Tumor Analysis (IOTA) predictive models, the logistic regression model 2 (LR2) and the IOTA Assessment of Different NEoplasias in the adneXa (ADNEX), in patients with ovarian masses and to compare their performance in preoperative discrimination between benign and malignant adnexal lesions. This was a retrospective diagnostic accuracy study with prospectively collected data, performed between January 2019 and December 2022, in a single tertiary gynecologic oncology center in Greece. The study included women with an adnexal lesion which underwent surgery within 6months after of using the LR2 and ADNEX protocol to assess the risk of malignancy. Correlation of the ultrasound findings with the postoperative histopathological analysis was performed. Receiver-operating characteristics (ROC) curve analysis was used to determine the diagnostic accuracy of the models to classify tumors; sensitivity and specificity were determined for each model and their performance was compared. Of the136 participants, 117 (86%) had benign ovarian masses and 19 (14%) had malignant tumors. The area under the ROC curve (AUC) of the LR2 model was 0.84 (95% CI 0.74-0.93), which was significantly higher than the AUC for ADNEX model: 0.78 (95% CI 0.67-0.89). At a cut off > 10%, the LR2 model had the highest sensitivity 89.5% (95% CI 66.9-98.7) and specificity 85.1% (95% CI 76.9-91.2) compared to ADNEX model [sensitivity 84.2% (95% CI 60.4-96.6) and specificity 71.8% (95% CI 62.7-79.7)]. IOTA LR2 had the highest accuracy in differentiating between benign and malignant ovarian masses. IOTA LR2 and ADNEX models were both useful tools in discriminating between benign and malignant ovarian masses.